[Paper Review] QuickEdit: Editing Text & Translations via Simple Delete Actions.

David Grangier, Michael Auli|arXiv (Cornell University)|Nov 13, 2017

Natural Language Processing Techniques10 citations

TL;DR

QuickEdit proposes a neural sequence-to-sequence framework for text and translation editing using simple delete actions: users mark tokens to be replaced, and the model generates a reformulated sentence avoiding those words. It achieves +11.4 BLEU on WMT-14 English-German translation post-editing (25.2 to 36.6), outperforming the baseline by +5.9 BLEU with minimal editing effort.

ABSTRACT

We propose a framework for computer-assisted text editing. It applies to translation post-editing and to paraphrasing and relies on very simple interactions: a human editor modifies a sentence by marking tokens they would like the system to change. Our model then generates a new sentence which reformulates the initial sentence by avoiding the words from the marked tokens. Our approach builds upon neural sequence-to-sequence modeling and introduces a neural network which takes as input a sentence along with deleted token markers. Our model is trained on translation bi-text by simulating post-edits. Our results on post-editing for machine translation and paraphrasing evaluate the performance of our approach. We show +11.4 BLEU with limited post-editing effort on the WMT-14 English-German translation task (25.2 to 36.6), which represents +5.9 BLEU over the post-editing baseline (30.7 to 36.6).

Motivation & Objective

• To reduce human post-editing effort in machine translation by enabling simple, targeted edits.
• To improve paraphrasing and translation quality through minimal user intervention.
• To develop a neural model that generates reformulated sentences by avoiding marked tokens.
• To train the model on simulated post-edits from parallel translation data.
• To evaluate the framework on both translation post-editing and paraphrasing tasks.

Proposed method

• The model uses a sequence-to-sequence architecture conditioned on input sentences with marked tokens to be deleted.
• It employs a neural network that takes both the original sentence and deletion markers as input.
• The model is trained on parallel monolingual and translation data by simulating post-edits through token deletion.
• Reformulation is achieved by generating a new sentence that avoids the deleted words while preserving meaning.
• The framework leverages attention mechanisms to focus on relevant context during generation.
• Training data is constructed by randomly masking tokens in source sentences and generating corresponding reformulations.

Experimental results

Research questions

• RQ1 Can simple delete actions significantly reduce post-editing effort in machine translation?
• RQ2 How effective is a neural model at generating meaningful reformulations after token deletion?
• RQ3 Does the proposed method improve BLEU scores compared to standard post-editing baselines?
• RQ4 Can the framework generalize to paraphrasing tasks beyond translation?
• RQ5 What is the trade-off between edit effort and output quality in text reformulation?

Key findings

• QuickEdit achieves a +11.4 BLEU improvement on the WMT-14 English-German translation task, rising from 25.2 to 36.6.
• It outperforms the post-editing baseline by +5.9 BLEU, which improved from 30.7 to 36.6.
• The method enables high-quality text reformulation with minimal human input, limited to marking tokens for deletion.
• The model generalizes effectively to both translation post-editing and paraphrasing tasks.
• The framework demonstrates strong performance gains using only simple, intuitive user interactions.
• The results confirm that targeted deletion-based editing is an effective and efficient approach for text generation and refinement.